From 95646f5cf8c3a995f0fb3c680f2d690b0172f193 Mon Sep 17 00:00:00 2001
From: "meiyou.hr" <meiyou.hr@alibaba-inc.com>
Date: Tue, 1 Apr 2025 19:29:41 +0800
Subject: [PATCH] add gp_binary test

---
 conf/functional/gp-test.conf |  0
 tests/gp-test/Readme.md      | 30 +++++++++++++++++
 tests/gp-test/get_ecos.sh    | 26 +++++++++++++++
 tests/gp-test/gpu_task.py    | 14 ++++++++
 tests/gp-test/install.sh     | 20 +++++++++++
 tests/gp-test/loop_inject.sh | 64 ++++++++++++++++++++++++++++++++++++
 tests/gp-test/parse.awk      | 17 ++++++++++
 tests/gp-test/run.sh         | 28 ++++++++++++++++
 8 files changed, 199 insertions(+)
 create mode 100644 conf/functional/gp-test.conf
 create mode 100644 tests/gp-test/Readme.md
 create mode 100644 tests/gp-test/get_ecos.sh
 create mode 100644 tests/gp-test/gpu_task.py
 create mode 100644 tests/gp-test/install.sh
 create mode 100644 tests/gp-test/loop_inject.sh
 create mode 100755 tests/gp-test/parse.awk
 create mode 100644 tests/gp-test/run.sh

diff --git a/conf/functional/gp-test.conf b/conf/functional/gp-test.conf
new file mode 100644
index 0000000..e69de29
diff --git a/tests/gp-test/Readme.md b/tests/gp-test/Readme.md
new file mode 100644
index 0000000..1d42dff
--- /dev/null
+++ b/tests/gp-test/Readme.md
@@ -0,0 +1,30 @@
+# gp-test
+## Description
+
+About the test suite
+
+## Homepage
+[http://](http://)
+
+## Version
+
+## Category
+
+functional/benchmarck/stress
+
+## Parameters
+
+- __value__ : about the value
+
+## Results
+
+```
+result: value
+```
+
+## Manual Run
+```
+step n: xxxx
+
+```
+
diff --git a/tests/gp-test/get_ecos.sh b/tests/gp-test/get_ecos.sh
new file mode 100644
index 0000000..07989cf
--- /dev/null
+++ b/tests/gp-test/get_ecos.sh
@@ -0,0 +1,26 @@
+#!/bin/bash
+
+if [[ "$1" == "" ]];then
+        channel="release"
+else
+        channel="$1"
+fi
+
+
+[[ -f latest ]] && rm -f latest
+wget https://gpu-profiling.oss-cn-hangzhou.aliyuncs.com/$channel/latest
+
+release=$(cat latest | head -n 1)
+checksum=$(cat latest | tail -n 1)
+ecos="ecos-ctl.$release"
+
+[[ -f $ecos ]] && rm -f $ecos
+
+echo "getting $ecos"
+wget https://gpu-profiling.oss-cn-hangzhou.aliyuncs.com/$channel/$ecos
+sha1=$(sha1sum $ecos | awk '{print $1}')
+if [[ "$checksum" == "$sha1" ]];then
+        echo "checked ok."
+else
+        echo "expect $checksum, but got $sha1"
+fi
\ No newline at end of file
diff --git a/tests/gp-test/gpu_task.py b/tests/gp-test/gpu_task.py
new file mode 100644
index 0000000..162e642
--- /dev/null
+++ b/tests/gp-test/gpu_task.py
@@ -0,0 +1,14 @@
+import torch
+import time
+
+def matrix_multiplication():
+    device = torch.device('cuda')
+    a = torch.randn(10000, 10000, device=device)
+    b = torch.randn(10000, 10000, device=device)
+    c = torch.matmul(a, b)
+    return c
+
+if __name__ == "__main__":
+    while True:
+        matrix_multiplication()
+        time.sleep(1)  # 模拟长时间运行
diff --git a/tests/gp-test/install.sh b/tests/gp-test/install.sh
new file mode 100644
index 0000000..254c86f
--- /dev/null
+++ b/tests/gp-test/install.sh
@@ -0,0 +1,20 @@
+#!/bin/bash
+# Avaliable environment:
+#
+# Download variable:
+# WEB_URL=
+# GIT_URL=
+
+
+# fetch()
+
+build()
+{
+    :
+}
+
+install()
+{
+     cp -Trf $TONE_BM_SUITE_DIR $TONE_BM_RUN_DIR
+}
+    
\ No newline at end of file
diff --git a/tests/gp-test/loop_inject.sh b/tests/gp-test/loop_inject.sh
new file mode 100644
index 0000000..6572801
--- /dev/null
+++ b/tests/gp-test/loop_inject.sh
@@ -0,0 +1,64 @@
+#!/bin/bash
+
+pass()
+{
+	echo "====PASS: $*"
+}
+
+fail()
+{
+	echo "====FAIL: $*"
+}
+# 默认值为1000
+CYCLE_COUNT=${CYCLE_COUNT:-1000}
+
+postfix=".$(cat latest | head -n 1)"
+ECOS="./ecos-ctl$postfix"
+
+if [ -n "$1" ]; then
+    if [ ! -f "$1" ]; then
+        echo "$1 not exist!"
+        exit 1
+    fi
+    ECOS=./$1
+fi
+
+chmod +x $ECOS
+echo "using $ECOS"
+
+for i in $(seq 1 $CYCLE_COUNT); do
+    echo "Current cycle: $i/$CYCLE_COUNT"
+
+    pids=$(nvidia-smi | grep python | awk '{print $5}' | tr '\n' ',')
+    if [ -z "$pids" ]; then
+        echo "error! No python processes found. exiting..."
+        fail "gp_loop_inject_$CYCLE_COUNT"
+        exit 1
+    fi
+
+    rm -f ./log/*
+    rm -f ./log/.cupti_fifo_*.buffer
+    pids="${pids%?}"
+    echo "inject to $pids"
+    $ECOS --debug monitor gpu probe --output ./log -d 1000 --pid $pids --force > ./log/ecoslog 2>&1 &
+    monitor_pid=$!
+
+    sleep 2
+    files=$(ls ./log/ | grep json | wc -l)
+    raws=$(ls ./log/ | grep raw | wc -l)
+
+    if [ "$files" -ne "$raws" ]; then
+        echo "error! File count mismatch. exiting..."
+        fail "gp_loop_inject_$CYCLE_COUNT"
+        kill -9 $monitor_pid
+        exit 1
+    fi
+done
+
+# End the monitored processes after testing is complete for this cycle.
+echo $pids | tr ',' '\n' | while read -r pid; do
+    kill -9 $pid
+done
+
+# End the ECOS monitoring process after all cycles are complete.
+pass "gp_loop_inject_$CYCLE_COUNT"
diff --git a/tests/gp-test/parse.awk b/tests/gp-test/parse.awk
new file mode 100755
index 0000000..784aa03
--- /dev/null
+++ b/tests/gp-test/parse.awk
@@ -0,0 +1,17 @@
+#!/usr/bin/awk -f
+
+/^====PASS:/ {
+        printf("%s: Pass\n",$2)
+}
+
+/^====FAIL:/ {
+        printf("%s: Fail\n",$2)
+}
+
+/^====SKIP:/ {
+        printf("%s: Skip\n",$2)
+}
+
+/^====WARN:/ {
+        printf("%s: Warning\n",$2)
+}
diff --git a/tests/gp-test/run.sh b/tests/gp-test/run.sh
new file mode 100644
index 0000000..6af325c
--- /dev/null
+++ b/tests/gp-test/run.sh
@@ -0,0 +1,28 @@
+#!/bin/bash
+# Avaliable environment:
+#
+# Download variable:
+# WEB_URL=
+# GIT_URL=
+
+
+run()
+{
+    export CYCLE_COUNT=10
+    mkdir -p log/
+    bash get_ecos.sh
+    docker run -d --name gptest_container --gpus all gptest_image:latest /bin/bash -c "while true; do sleep 60; done"
+    docker exec -d gptest_container bash -c "python3 /root/gpu_task.py"
+    sleep 5
+    bash loop_inject.sh
+    
+    docker rm -f gptest_container
+    upload_archives log/
+
+}
+
+parse()
+{
+    $TONE_BM_SUITE_DIR/parse.awk
+}
+    
\ No newline at end of file
-- 
Gitee